xref: /openbsd-src/usr.bin/rsync/socket.c (revision 9f11ffb7133c203312a01e4b986886bc88c7d74b)
1 /*	$Id: socket.c,v 1.6 2019/02/12 19:20:15 benno Exp $ */
2 /*
3  * Copyright (c) 2019 Kristaps Dzonsons <kristaps@bsd.lv>
4  *
5  * Permission to use, copy, modify, and distribute this software for any
6  * purpose with or without fee is hereby granted, provided that the above
7  * copyright notice and this permission notice appear in all copies.
8  *
9  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
10  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
11  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
12  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
13  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
14  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
15  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
16  */
17 #include <sys/stat.h>
18 #include <sys/socket.h>
19 #include <arpa/inet.h>
20 #include <netinet/in.h>
21 
22 #include <assert.h>
23 #include <ctype.h>
24 #include <errno.h>
25 #include <fcntl.h>
26 #include <inttypes.h>
27 #include <netdb.h>
28 #include <poll.h>
29 #include <resolv.h>
30 #include <stdlib.h>
31 #include <string.h>
32 #include <unistd.h>
33 
34 #include "extern.h"
35 
36 /*
37  * Defines a resolved IP address for the host
38  * There can be many, IPV4 or IPV6.
39  */
40 struct	source {
41 	int		 family; /* PF_INET or PF_INET6 */
42 	char		 ip[INET6_ADDRSTRLEN]; /* formatted string */
43 	struct sockaddr_storage sa; /* socket */
44 	socklen_t	 salen; /* length of socket buffer */
45 };
46 
47 /*
48  * Connect to an IP address representing a host.
49  * Return <0 on failure, 0 on try another address, >0 on success.
50  */
51 static int
52 inet_connect(struct sess *sess, int *sd,
53 	const struct source *src, const char *host)
54 {
55 	int	 c, flags;
56 
57 	if (*sd != -1)
58 		close(*sd);
59 
60 	LOG2(sess, "trying: %s, %s", src->ip, host);
61 
62 	if ((*sd = socket(src->family, SOCK_STREAM, 0)) == -1) {
63 		ERR(sess, "socket");
64 		return -1;
65 	}
66 
67 	/*
68 	 * Initiate blocking connection.
69 	 * We use the blocking connect() instead of passing NONBLOCK to
70 	 * the socket() function because we don't need to do anything
71 	 * while waiting for this to finish.
72 	 */
73 
74 	c = connect(*sd,
75 		(const struct sockaddr *)&src->sa,
76 		src->salen);
77 	if (c == -1) {
78 		if (errno == ECONNREFUSED || errno == EHOSTUNREACH) {
79 			WARNX(sess, "connect refused: "
80 				"%s, %s", src->ip, host);
81 			return 0;
82 		}
83 		ERR(sess, "connect");
84 		return -1;
85 	}
86 
87 	/* Set up non-blocking mode. */
88 
89 	if ((flags = fcntl(*sd, F_GETFL, 0)) == -1) {
90 		ERR(sess, "fcntl");
91 		return -1;
92 	} else if (fcntl(*sd, F_SETFL, flags|O_NONBLOCK) == -1) {
93 		ERR(sess, "fcntl");
94 		return -1;
95 	}
96 
97 	return 1;
98 }
99 
100 /*
101  * Resolve the socket addresses for host, both in IPV4 and IPV6.
102  * Once completed, the "dns" pledge may be dropped.
103  * Returns the addresses on success, NULL on failure (sz is always zero,
104  * in this case).
105  */
106 static struct source *
107 inet_resolve(struct sess *sess, const char *host, size_t *sz)
108 {
109 	struct addrinfo	 hints, *res0, *res;
110 	struct sockaddr	*sa;
111 	struct source	*src = NULL;
112 	size_t		 i, srcsz = 0;
113 	int		 error;
114 
115 	*sz = 0;
116 
117 	memset(&hints, 0, sizeof(hints));
118 	hints.ai_family = PF_UNSPEC;
119 	hints.ai_socktype = SOCK_DGRAM; /* DUMMY */
120 
121 	error = getaddrinfo(host, "873", &hints, &res0);
122 
123 	LOG2(sess, "resolving: %s", host);
124 
125 	if (error == EAI_AGAIN || error == EAI_NONAME) {
126 		ERRX(sess, "DNS resolve error: %s: %s",
127 			host, gai_strerror(error));
128 		return NULL;
129 	} else if (error) {
130 		ERRX(sess, "DNS parse error: %s: %s",
131 			host, gai_strerror(error));
132 		return NULL;
133 	}
134 
135 	/* Allocate for all available addresses. */
136 
137 	for (res = res0; NULL != res; res = res->ai_next)
138 		if (res->ai_family == AF_INET ||
139 		    res->ai_family == AF_INET6)
140 			srcsz++;
141 
142 	if (srcsz == 0) {
143 		ERRX(sess, "no addresses resolved: %s", host);
144 		freeaddrinfo(res0);
145 		return NULL;
146 	}
147 
148 	src = calloc(srcsz, sizeof(struct source));
149 	if (src == NULL) {
150 		ERRX(sess, "calloc");
151 		freeaddrinfo(res0);
152 		return NULL;
153 	}
154 
155 	for (i = 0, res = res0; NULL != res; res = res->ai_next) {
156 		if (res->ai_family != AF_INET &&
157 		    res->ai_family != AF_INET6)
158 			continue;
159 
160 		assert(i < srcsz);
161 
162 		/* Copy the socket address. */
163 
164 		src[i].salen = res->ai_addrlen;
165 		memcpy(&src[i].sa, res->ai_addr, src[i].salen);
166 
167 		/* Format as a string, too. */
168 
169 		sa = res->ai_addr;
170 		if (res->ai_family == AF_INET) {
171 			src[i].family = PF_INET;
172 			inet_ntop(AF_INET,
173 				&(((struct sockaddr_in *)sa)->sin_addr),
174 				src[i].ip, INET6_ADDRSTRLEN);
175 		} else {
176 			src[i].family = PF_INET6;
177 			inet_ntop(AF_INET6,
178 				&(((struct sockaddr_in6 *)sa)->sin6_addr),
179 				src[i].ip, INET6_ADDRSTRLEN);
180 		}
181 
182 		LOG2(sess, "DNS resolved: %s: %s", host, src[i].ip);
183 		i++;
184 	}
185 
186 	freeaddrinfo(res0);
187 	*sz = srcsz;
188 	return src;
189 }
190 
191 /*
192  * Process an rsyncd preamble line.
193  * This is either free-form text or @RSYNCD commands.
194  * Return <0 on failure, 0 on try more lines, >0 on finished.
195  */
196 static int
197 protocol_line(struct sess *sess, const char *host, const char *cp)
198 {
199 	int	major, minor;
200 
201 	if (strncmp(cp, "@RSYNCD: ", 9)) {
202 		LOG0(sess, "%s", cp);
203 		return 0;
204 	}
205 
206 	cp += 9;
207 	while (isspace((unsigned char)*cp))
208 		cp++;
209 
210 	/* @RSYNCD: OK indicates that we're finished. */
211 
212 	if (strcmp(cp, "OK") == 0)
213 		return 1;
214 
215 	/*
216 	 * Otherwise, all we have left is our version.
217 	 * There are two formats: x.y (w/submodule) and x.
218 	 */
219 
220 	if (sscanf(cp, "%d.%d", &major, &minor) == 2) {
221 		sess->rver = major;
222 		return 0;
223 	} else if (sscanf(cp, "%d", &major) == 1) {
224 		sess->rver = major;
225 		return 0;
226 	}
227 
228 	ERRX(sess, "rsyncd protocol error: unknown command");
229 	return -1;
230 }
231 
232 /*
233  * Pledges: dns, inet, unveil, rpath, cpath, wpath, stdio, fattr.
234  *
235  * Pledges (dry-run): -cpath, -wpath, -fattr.
236  * Pledges (!preserve_times): -fattr.
237  */
238 int
239 rsync_socket(const struct opts *opts, const struct fargs *f)
240 {
241 	struct sess	  sess;
242 	struct source	 *src = NULL;
243 	size_t		  i, srcsz = 0;
244 	int		  sd = -1, rc = 0, c;
245 	char		**args, buf[BUFSIZ];
246 	uint8_t		  byte;
247 
248 	memset(&sess, 0, sizeof(struct sess));
249 	sess.lver = RSYNC_PROTOCOL;
250 	sess.opts = opts;
251 
252 	assert(f->host != NULL);
253 	assert(f->module != NULL);
254 
255 	if ((args = fargs_cmdline(&sess, f)) == NULL) {
256 		ERRX1(&sess, "fargs_cmdline");
257 		return 0;
258 	}
259 
260 	/* Resolve all IP addresses from the host. */
261 
262 	if ((src = inet_resolve(&sess, f->host, &srcsz)) == NULL) {
263 		ERRX1(&sess, "inet_resolve");
264 		free(args);
265 		return 0;
266 	}
267 
268 	/* Drop the DNS pledge. */
269 
270 	if (pledge("stdio rpath wpath cpath fattr getpw inet unveil", NULL) == -1) {
271 		ERR(&sess, "pledge");
272 		goto out;
273 	}
274 
275 	/*
276 	 * Iterate over all addresses, trying to connect.
277 	 * When we succeed, then continue using the connected socket.
278 	 */
279 
280 	assert(srcsz);
281 	for (i = 0; i < srcsz; i++) {
282 		c = inet_connect(&sess, &sd, &src[i], f->host);
283 		if (c < 0) {
284 			ERRX1(&sess, "inet_connect");
285 			goto out;
286 		} else if (c > 0)
287 			break;
288 	}
289 
290 	/* Drop the inet pledge. */
291 	if (pledge("stdio rpath wpath cpath fattr getpw unveil", NULL) == -1) {
292 		ERR(&sess, "pledge");
293 		goto out;
294 	}
295 
296 	if (i == srcsz) {
297 		ERRX(&sess, "cannot connect to host: %s", f->host);
298 		goto out;
299 	}
300 
301 	/* Initiate with the rsyncd version and module request. */
302 
303 	LOG2(&sess, "connected: %s, %s", src[i].ip, f->host);
304 
305 	(void)snprintf(buf, sizeof(buf), "@RSYNCD: %d", sess.lver);
306 	if (!io_write_line(&sess, sd, buf)) {
307 		ERRX1(&sess, "io_write_line");
308 		goto out;
309 	}
310 
311 	LOG2(&sess, "requesting module: %s, %s", f->module, f->host);
312 
313 	if (!io_write_line(&sess, sd, f->module)) {
314 		ERRX1(&sess, "io_write_line");
315 		goto out;
316 	}
317 
318 	/*
319 	 * Now we read the server's response, byte-by-byte, one newline
320 	 * terminated at a time, limited to BUFSIZ line length.
321 	 * For this protocol version, this consists of either @RSYNCD
322 	 * followed by some text (just "ok" and the remote version) or
323 	 * the message of the day.
324 	 */
325 
326 	for (;;) {
327 		for (i = 0; i < sizeof(buf); i++) {
328 			if (!io_read_byte(&sess, sd, &byte)) {
329 				ERRX1(&sess, "io_read_byte");
330 				goto out;
331 			}
332 			if ((buf[i] = byte) == '\n')
333 				break;
334 		}
335 		if (i == sizeof(buf)) {
336 			ERRX(&sess, "line buffer overrun");
337 			goto out;
338 		} else if (i == 0)
339 			continue;
340 
341 		/*
342 		 * The rsyncd protocol isn't very clear as to whether we
343 		 * get a CRLF or not: I don't actually see this being
344 		 * transmitted over the wire.
345 		 */
346 
347 		assert(i > 0);
348 		buf[i] = '\0';
349 		if (buf[i - 1] == '\r')
350 			buf[i - 1] = '\0';
351 
352 		if ((c = protocol_line(&sess, f->host, buf)) < 0) {
353 			ERRX1(&sess, "protocol_line");
354 			goto out;
355 		} else if (c > 0)
356 			break;
357 	}
358 
359 	/*
360 	 * Now we've exchanged all of our protocol information.
361 	 * We want to send our command-line arguments over the wire,
362 	 * each with a newline termination.
363 	 * Use the same arguments when invoking the server, but leave
364 	 * off the binary name(s).
365 	 * Emit a standalone newline afterward.
366 	 */
367 
368 	if (f->mode == FARGS_RECEIVER || f->mode == FARGS_SENDER)
369 		i = 3; /* ssh host rsync... */
370 	else
371 		i = 1; /* rsync... */
372 
373 	for ( ; NULL != args[i]; i++)
374 		if (!io_write_line(&sess, sd, args[i])) {
375 			ERRX1(&sess, "io_write_line");
376 			goto out;
377 		}
378 	if (!io_write_byte(&sess, sd, '\n')) {
379 		ERRX1(&sess, "io_write_line");
380 		goto out;
381 	}
382 
383 	/*
384 	 * All data after this point is going to be multiplexed, so turn
385 	 * on the multiplexer for our reads and writes.
386 	 */
387 
388 	/* Protocol exchange: get the random seed. */
389 
390 	if (!io_read_int(&sess, sd, &sess.seed)) {
391 		ERRX1(&sess, "io_read_int");
392 		goto out;
393 	}
394 
395 	/* Now we've completed the handshake. */
396 
397 	if (sess.rver < sess.lver) {
398 		ERRX(&sess, "remote protocol is older "
399 			"than our own (%" PRId32 " < %" PRId32 "): "
400 			"this is not supported",
401 			sess.rver, sess.lver);
402 		goto out;
403 	}
404 
405 	sess.mplex_reads = 1;
406 	LOG2(&sess, "read multiplexing enabled");
407 
408 	LOG2(&sess, "socket detected client version %" PRId32
409 		", server version %" PRId32 ", seed %" PRId32,
410 		sess.lver, sess.rver, sess.seed);
411 
412 	assert(f->mode == FARGS_RECEIVER);
413 
414 	LOG2(&sess, "client starting receiver: %s", f->host);
415 	if (!rsync_receiver(&sess, sd, sd, f->sink)) {
416 		ERRX1(&sess, "rsync_receiver");
417 		goto out;
418 	}
419 
420 #if 0
421 	/* Probably the EOF. */
422 	if (io_read_check(&sess, sd))
423 		WARNX(&sess, "data remains in read pipe");
424 #endif
425 
426 	rc = 1;
427 out:
428 	free(src);
429 	free(args);
430 	if (sd != -1)
431 		close(sd);
432 	return rc;
433 }
434